upload local tensorboard logs to clearml server

[XZLancer]

Proposal Summary

Hi ClearML team, I have some local tensorboard logs in previous training, I‘d like to upload them to the cleaml server so that I can have a unified management and it is convenient for comparison. Is there a suitable way to achieve this? Thanks for your time and looking forward to your reply!

Motivation

Upload local tensorboard logs to clearml server for unified management and convenience checking

[eugen-ajechiloae-clearml]

Hi @XZLancer ! You could create a task for each previous training run you have logs for. If the logs contain scalars or plots, you could parse them and report them using https://clear.ml/docs/latest/docs/references/sdk/logger#report_scalar or other appropriate reporting functions from the clearml logger. Text logs can be reported to the experiment console logs with https://clear.ml/docs/latest/docs/references/sdk/logger#report_text.
Tasks can then be compared using the compare functionality in the UI. Reference: https://clear.ml/docs/latest/docs/webapp/webapp_exp_comparing/

[XZLancer]

I tried the following code snippet:

# Initialize TensorBoard EventAccumulator to read the logs
event_accumulator = EventAccumulator(tensorboard_log_dir)
event_accumulator.Reload()  # Load the events

# Get scalar tags (metrics like loss, accuracy, etc.)
scalar_tags = event_accumulator.Tags()["scalars"]

# Report scalar metrics to ClearML
for tag in scalar_tags:
    scalar_events = event_accumulator.Scalars(tag)
    for event in scalar_events:
        task.logger.report_scalar(
            title=tag,
            series=tag,
            value=event.value,
            iteration=event.step,
        )

and I found that it runs very slowly(it takes about 30 minutes to report the 25k steps of scalars). Are there any reasons or solution? Thanks for reply!
EDIT:
The program keeps reporting

Retrying (Retry(total=237, connect=240, read=237, redirect=240, status=240)) after connection broken by 'ReadTimeoutError("HTTPConnectionPool(host='192.168.104.2', port=8010): Read timed out. (read timeout=300.0)")': /v2.13/events.add_batch

and the scalars are displayed much slower in the webUI than the program reports.